Another possibility in case of bounded variables is to use a polytopic model, i.e., if x belongs to [a, b] and y belongs to [c, d], then the function xy is mapped to a polytope that is a linear combination whose coefficients depend on the mapped vertexes.
Truncation of the Taylor series expansion around certain point gives a linear model. This model is different from the polytope model that is also linear. The first one is a good approximation around the given point and the second one leads to linear combinations inside the polytope.